Amazon (NASDAQ:AMZN) could start developing specialized AI chips, according to The Information. The move wouldn't be surprising, since Amazon previously expanded its chipmaking abilities with its acquisitions of Israeli chipmaker Annapurna Labs in 2015 and security camera maker Blink last year.
The report claims that Amazon is developing new AI chips for new Echo devices, which could potentially improve their offline performance. It could also produce new AI chips for its Amazon Web Services (AWS) data centers.
Amazon's investments in chips closely mirror similar moves by Apple (NASDAQ:AAPL) and Alphabet's (NASDAQ:GOOG) (NASDAQ:GOOGL) Google. Apple has been gradually replacing third-party chips in its iPhones with its own chips, cutting suppliers like Imagination Technologies and Dialog Semiconductor out of the loop.
Google created a custom processor called the Tensor Processing Unit (TPU) several years ago for its own machine-learning tasks. Over the past two years, Google developed custom AI chips for devices like its Clips Camera, as well as the image processor for its Pixel 2 smartphone.
Amazon, Google, and Apple's chipmaking moves don't really affect each other, since they're producing chips for their own hardware. But they could hurt Intel (NASDAQ:INTC) and NVIDIA (NASDAQ:NVDA) -- which both sell large quantities of chips to data centers.
How Amazon could hurt Intel
Intel's Xeon CPUs are installed in 99% of data centers across the world. Its data center group revenues, which accounted for 30% of its top line, rose 11% to $19.1 billion in 2017. Intel is also expanding into adjacent markets -- like chips for cars, wearables, drones, and Internet of Things (IoT) devices -- to diversify its business away from PCs and data centers.
However, AWS is the largest cloud infrastructure platform in the world and currently runs across 53 availability zones. Each availability zone is comprised of multiple data centers. The servers in these centers process AWS tasks with Xeon processors.
But for machine-learning tasks, Amazon installs NVIDIA's high-end Tesla GPUs alongside Intel's Xeons. Intel launched new Xeon CPUs for machine-learning purposes to counter NVIDIA's growth in the AI market, but NVIDIA claims Intel's stand-alone CPUs still can't match the machine-learning horsepower of its high-end GPUs.
If Amazon ramps up its chipmaking efforts, it could add custom AI chips to its data centers instead of upgrading its existing Xeons. It could also sell its custom AI chips for Echo to third-party hardware makers that are producing Alexa-powered devices. That would disrupt Intel's plans to help companies create Alexa-powered devices with its "Speech Enabling Developer Kit" -- which it introduced last year to strengthen its IoT business.
How Amazon could hurt NVIDIA
Amazon could gradually favor its own AI chips over NVIDIA's GPUs in its own data centers. Google did the same thing with its TPUs and claims that its own chips offer better performance than NVIDIA's GPUs -- a claim NVIDIA disputes.
NVIDIA's data center chip revenue more than doubled annually to $606 million last quarter and accounted for 21% of its top line. That makes it NVIDIA's fastest-growing business, and its future growth could offset a cyclical slowdown in gaming GPU sales or tougher competition in the automotive chip market.
However, NVIDIA already faces competition in the data center market from Intel and Google. Intel's development of its own discrete GPUs (which it might pair with its own Xeons for machine learning) could cause more headaches. Therefore, Amazon's entrance into the AI chip market could become another unpredictable headwind for NVIDIA.
But let's not get ahead of ourselves...
Amazon's interest in custom AI chips is worth watching, but investors shouldn't expect these chips to disrupt Intel and NVIDIA's dominance of the server CPU and machine-learning GPU markets anytime soon.
Intel and NVIDIA are still the "best in breed" leaders of those markets, so Amazon will likely install its own chips alongside those chips instead of replacing them. But over the long term, Amazon could gradually reduce its dependence on Intel and NVIDIA to tighten its control over its own ecosystem.